mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-11 03:38:38 +00:00
3ee6a70d04
Adds QuickSilver Pro (MachineFi Inc., https://quicksilverpro.io) as a JSON-configured OpenAI-compatible provider, following the SCX.ai provider pattern (PR #34752). - litellm/llms/openai_like/providers.json: `quicksilverpro` entry with base_url https://api.quicksilverpro.io/v1, env vars QUICKSILVERPRO_API_KEY / QUICKSILVERPRO_API_BASE, param_mappings {max_completion_tokens -> max_tokens}, and supported endpoints /v1/chat/completions, /v1/responses, /v1/messages (Claude models). - constants.py: appended to openai_compatible_endpoints and openai_compatible_providers. - types/utils.py: LlmProviders.QUICKSILVERPRO. - model_prices_and_context_window.json (+ backup, kept in sync): 8 models with prices from https://quicksilverpro.io/pricing.json (catalog_version 40, updated 2026-09-30), USD per 1M tokens converted to USD per token: claude-opus-5-5 (1.60/8.00), claude-sonnet-5-5 (1.00/5.00), gpt-6.1-sol (2.00/10.00), deepseek-v4.1-flash (0.125/0.55), glm-5.3-flash (0.06/0.20), mimo-v2.6-pro (0.348/0.696), gemini-3.8-flash (0.6375/3.1875), qwen3.8-27b (0.34/2.04), with cache-read/cache-write prices and context windows from the same catalog. Where pricing.json lists no output cap (deepseek-v4.1-flash, mimo-v2.6-pro), the legacy max_tokens falls back to the context window. - provider_endpoints_support.json (+ backup): chat_completions, messages and responses enabled; embeddings, image_generations, audio_transcriptions, audio_speech, moderations, batches, rerank and a2a disabled (no JSON-provider mechanism exists for images). - proxy/public_endpoints/provider_create_fields.json: QUICKSILVERPRO form entry (required api_key, optional api_base with default placeholder), default model quicksilverpro/claude-sonnet-5-5. - UI: provider enum + slug + logo + placeholder maps (provider_info_helpers.tsx), plate logo treatment (logoTreatments.ts), quicksilverpro.svg in public assets and the built proxy assets. - tests/unit/llms/openai_like/test_quicksilverpro_provider.py (25 tests): registry/config, provider resolution + api_base override, param mapping, router config, dashboard registration, and per-model behaviour checks that charge/read each entry's own prices via cost_per_token + get_model_info instead of pinning upstream numbers. respx-mocked (no network) request tests cover all three declared endpoints: chat/completions (with max_completion_tokens -> max_tokens mapping asserted on the wire), /v1/responses, and /v1/messages. Local validation: 290 passed (tests/unit/llms/openai_like/ + tests/unit/test_model_prices_schema.py), 90 UI tests passed (provider_info_helpers.test.tsx, logoTreatments.test.ts) with no type errors. |
||
|---|---|---|
| .. | ||
| litellm-dashboard | ||
| Dockerfile | ||
| nginx.conf | ||